Image Recognition With Sipeed MaiX and Arduino IDE/Micropython
HTML-код
- Опубликовано: 14 июл 2024
- Recognize anything! Well, almost.
How to create your own custom image classifier with transfer learning in Keras, convert the trained model to .kmodel format and run it on Sipeed board (can be any board, Bit/Dock or Go) using Micropython or Arduino IDE
The board for this video was kindly provided by Seeed studio, check out this board and other hardware for machine learning acceleration at Seeed studio store!
www.seeedstudio.com/Sipeed-MA...
Link to the tutorial
www.instructables.com/id/Tran...
Link to full length demo:
• Image Recognition With...
Add me on LinkedIn if you have any question:
/ dmitry-maslov-ai
Credits for music:
• Beyond - A Chillwave Mix
From Beyond - A Chillwave Mix (Kattch - Sweet Jam) 
Наука
![XO (Only If You Say Yes) - ENHYPEN エンハイプン 엔하이픈 [Music Bank] | KBS WORLD TV 240712](http://i.ytimg.com/vi/wBT1bKKwPTo/mqdefault.jpg)
nice video. just ordered my dan dock and it's still on the way
Glad to hear you liked the video!
Dan dock is supposed to work exactly the same, except it doesn't have an SD card, so you'll need to upload the model into the flash memory.
great tutorial - thank you! I have such a board here for months now but I cannot do much with it because I have only a Mac and kflash does not work with it. I will try Maxduino or platform.io later - I might have more luck with it ;)
Thank you! I think latest version of kflash.py supports MacOS, you can download it here github.com/kendryte/kflash.py
macOS Requirements
# Install Homebrew, an awesome package manager for macOS
/usr/bin/ruby -e "$(curl -fsSL raw.githubusercontent.com/Homebrew/install/master/install)"
brew install python
python3 -m pip3 install pyserial
python3 -m pip3 install pyelftools
thank you, go ahead
I will :)
Thank you ............
My pleasure :)
hello i have a problem when i convert of .h5 to .tflite with ''from tensorflow.contrib import lite
make sure you have proper versions of tensorflow and keras installed! can you check you tf and keras versions?
Hey dimitri, I am following your tutorial from github of the classification model to run it on Colab to use it on K210. However, there's some issues with the compatibility of the Adam optimizer function and TensorFlow version 2.x, . Colab no longer works with TensorFlow 1.x, so could you help me move forward?
After a while, I moved all the K210 scripts to this repository github.com/AIWintermuteAI/aXeleRate. Have you found it? It uses tf 2.x.
Nice video.
I am using the Arduino development environment for my project. I am testing the camera module using the code provided on the website/Arduino to capture the image and display on LCD. All I am getting on the Serial monitor is snap fail and LCD remains RED forever.
I can't find the solution anywhere. Is there something I am missing?
Any help will be appreciated. Thank you.
First of all, Arduino IDE for maix boards is poorly supported and seems to be about 6 months behind Micropython firmware. Which is a shame of course, since C code is superior in speed.
I'm not familiar with this error message. Sensor reset fail or snapshot fail message means the module cannot get the image from camera (camera not found), Usually it is a connection problem - camera connector broken or inserted the wrong way.
Hey, I am working on Monocular Depth Estimation on k210 specifically for human detection. Does the board capable of running the model. One of my models pretrained on NYU dataset has a size of 164 MB. I don't know would that be suitable for the device. Any Help is appreciated. Thanks
Hi there! Have a look at the documentation for K210 model converter here
github.com/kendryte/nncase/blob/master/docs/FAQ_EN.md
Short answer - no, 164 is waaaay too big. The largest model size can be about 12-15 MB(before quantization) and this is somewhat stretching it.
Very interesting topic though, Monocular Depth Estimation. Have you considered using smaller models? What do you use for encoder part of the model(I only worked with image segmentation models, I;m assuming depth estimation models are similar - encoder > feature maps > upsample to output layer mask)
@@Hardwareai Can stereo depth estimation be possible through MAix Binocular Camera which offers Depth Vision?
I tried monocular depth but smaller models are still bigger for the chip. If this stereo camera can be applicable for estimation then that will be great help
Thanks again
Hello, and thanks a lot for your videos. Can the Sipeed board detect and count multiple objects ? Thanks a lot !
Detecting - yes, for sure. I was traveling last two weeks, but I have a new video planned to publish next week about object detection models with K210. Now, counting is another topic at all, since for that you will need to do some extra operations outside of the neural network. Some functions in OpenMV can be useful for that.
What's the use case?
@@Hardwareai Thanks a lot for the answer. I would like to be able to count bees in real time, getting in / out of their hive.
ruclips.net/video/87c6dCgXeJo/видео.html Have a look at my new tutorial on object detection. Although I have to say, detecting and counting bees while possible with K210, but challenging. The limitation is maximum resolution(320x240) of the image, which makes detecting small objects difficult. I think you can do it, if you place the camera into hive entrance and add some light source
@@Hardwareai Already seen it, very useful, thanks a lot, that is what I need. I think it could work because I do not need the exact number of bees getting in / out, just a ratio, so I can put the camera close to them. I need to do some test with the hardware for sure, it is on the way ! Thanks again for your great videos !
Hy I have problem re flashing K210 Maix Pi board [ERROR] No vaild Kendryte K120 found
Well, without more information it is difficult to assist you... Have you chosen the correct ype of the board in kflash? Can you connect to board using MaixPy IDE or serial communication?
Awesome video bro, could you please guide me on how can I run openCV on Sipeed m1 dock!
openCV is too heavy to run on embedded system like MAIX I boards. there is OpenMV available within Micropython firmware though - you don't need anything special in order to use it. Just download latest firmware and use OpenMV functions.
can you do Grove AI HAT for Edge computing too? I wanna know how will it help with robotic arm to pick up specific object..
Oh, well.... That product is not under active development now. From hardware standpoint it is great idea, but it still lacks proper communication interface.
@@Hardwareai which one you recommend?
Picking up specific object with robotic arm. Are you using ROS? If you do not require anything too fancy (like real time obstacle avoidance), then just Raspberry Pi 4 might do - no need for accelerators if you do not require the speed. If you do require the speed, then how about Google Coral USB accelerator. It is more expensive, but has much better integration with the host. Or you need a standalone solution(meaning you will run both object detection and robotic arm control from the same board and you want that board to be as cheap as possible)?
@@Hardwareai i was thinking of making harvesting cili using robotic arm. So:
1. Need to identify ripe red cili(target)
2. Ability to detect obstacle covering target i.e leaf
3. The harvested cili will be carried by the robot tank(?) then to a bigger containment, so obstacle avoidance will be put to the test i.e rocks and big branch
Its still in early stage. I did consider using ROS but my brain hurts and I dunno if ROS is a right path (if you say it is I'm gonna invest my time in it).
I am using Raspberry pi 3B+ for robotic arm+opencv and arduino uno for tank motor control+other sensor. I thought Grove AI HAT can help accelerate things tho didnt know they're lack in support(thats me being cheap) lol
The project sounds great. I think it definitely makes sense for you to switch to R Pi 4 or Jetson Nano - you have a lot of things going on and for some of them you need real-time processing.
Yes, ROS is the right path - there is a lot of integration work that needs to be done and you'll have much harder time without ROS(or other middleware for your components).
You first challenge is grasping the red chili - that would require object detection, inverse kinematics and grasp planning.
Then it's navigation on rough terrain - will need path planning and some form of SLAM(for example V-SLAM).
Does anyone know why i can't acess the bbs.sipeed.com ?? I'm trying to follow the discussions there about sipeed maix go but there is always an error when loading the page. Is it Offline? I've tried with some proxy but it didn't work. I'm from Brazil. Please someone help me :S
Hello! I checked just now. I can access it, using proxy in Japan. Can you change the network, for example try from the mobile?
@@Hardwareai Thank you for the answer, my friend! The next day, it worked fine. I think it was out but then got back.
Now, my main problem is understanding how to use it... I bought it recently and im still a noob hahahaha
I can't find much documentation... Even on the site, there ain't much explanaiton, just a few examples. Where do you get everything you need to program it? I think you know Chinese, by the t-shirt, and it seems to me that most of the info is in chinese hahaha maybe im wrong and really just a noob
Sorry for any english mistake, cause im brazilian :)
@@gauninha12 Your English is great, no worse than mine, haha.
Hm, there are all sorts of demos in this repo:
github.com/sipeed/MaixPy_scripts
they are mostly related to using network, GPIO, screen and so on.
For object detection and recognition you can have a look at my articles on instructables.
There is some info in Maixpy Github(in issues as well).
Overall, this is one of the most user-friendly Chinese boards xD
@@Hardwareai Thank you again, my friend! I have already done some examples, but when i try to download models or upgrade the firmware using the Kflash tool, it tells me that there is no K210 chip found, but using the IDE, it works properly. I am trying to search its driver on the internet, but i can't find. Do you need any driver to make kflash recognize the chip using windows?
If it works properly using IDE, then it's not a driver problem :) There might be a problem with Kflashtool permissions. What OS are you on?
🖖🤓
🖖
Gostei muito desta placa, ja tenho a minha. Gostaria de treina la para detectar veiculos , trucks carretas, carros, motos.
Sorry, no speak Spanish 0_o